Building a Best-in-Class Automated De-identification Tool for Electronic Health Records Through Ensemble Learning

The natural language portions of electronic health records (EHRs) communicate critical information about disease and treatment progression. However, the presence of personally identifiable information (PII) in this data constrains its broad reuse. Despite continuous improvements in methods for the automated detection of PII, the presence of residual identifiers in clinical notes requires manual validation and correction. However, manual intervention is not a scalable solution for large EHR datasets. Here, we describe an automated de-identification system that employs an ensemble architecture, incorporating attention-based deep learning models and rule-based methods, supported by heuristics for detecting PII in EHR data. Upon detection of PII, the system transforms these detected identifiers into plausible, though fictional, surrogates to further obfuscate any leaked identifier. We evaluated the system with a publicly available dataset of 515 notes from the I2B2 2014 de-identification challenge and a dataset of 10,000 notes from the Mayo Clinic. In comparison with other existing tools considered best-in-class, our approach outperforms them with a recall of 0.992 and 0.994 and a precision of 0.979 and 0.967 on the I2B2 and the Mayo Clinic data, respectively. The automated de-identification system presented here can enable the generation of de-identified patient data at the scale required for modern machine learning applications to help accelerate medical discoveries.